Soumith Chintala
Member Of Technical Staff at K-Scale Labs
New York, New York, United States
Summary
🤩
Rockstar🎓
Top SchoolSoumith Chintala is a veteran AI researcher-engineer with 15+ years building core machine learning infrastructure and production-grade research systems, currently a Member of Technical Staff in New York. He led PyTorch from its early days to industry-wide adoption, driving engineering, product and partnerships for large cross-industry deployments while maintaining deep hands‑on contributions to CUDA kernels, CuDNN bindings and performance-critical backends. Equally fluent in research and engineering, his papers on GANs (DCGAN, WGAN) and high-impact open-source work (torch, torchvision, torchaudio, convnet-benchmarks, pytorch examples/tutorials) reflect both academic rigor and practical impact measured in millions of downloads. He’s also an active angel investor and advisor across dozens of AI startups, bringing operator experience from infra to product. A practical systems thinker, he often surfaces subtle implementation fixes—like stride/non-contiguous tensor bug fixes and deterministic robotics transforms—that quietly increase reliability at scale.
16 years of coding experience
11 years of employment as a software developer
Master of Science Computer Science, Master of Science Computer Science at New York UniversityBachelor of Technology Information Technology, Bachelor of Technology Information Technology at VIT University
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (95)
website-development10
operation10
python10
image-processing10
testing10
bash10
cmake10
c1110
datasets10
c1710
lua10
data-loading10
automation10
deep-learning10
gpu10
Programming languages (24)
Github contributions (5)
torch/cunn
Aug 2012 - Sep 2017
Role in this project:
Back-end Developer
Back-end DeveloperContributions:373 commits, 323 PRs, 313 pushes in 5 years 1 month
Contributions summary:Soumith contributed significantly to the `torch/cunn` repository, focusing on implementing and improving CUDA kernels for neural network operations. The commits involved the addition and modification of core functionalities within the library, including creating a CUDA-based implementation of SpatialConvolutionMap, fixing bugs in existing kernels like SpatialConvolutionMM and SpatialMaxPooling, and adding new kernels like SoftMax. These changes demonstrate a deep understanding of CUDA programming and neural network implementation.
soumith/cudnn.torch
Sep 2014 - Aug 2017
Torch-7 FFI bindings for NVIDIA CuDNN
Role in this project:Back-end Developer
Back-end DeveloperContributions:271 commits, 181 PRs, 224 pushes in 2 years 11 months
Contributions summary:Soumith primarily contributed to the core functionality of the `cudnn.torch` library, focusing on wrapping and interfacing with NVIDIA's CuDNN library for deep learning. Their commits involved defining C FFI bindings, implementing core modules like SpatialConvolution, and addressing bug fixes related to the integration of CuDNN functions within the Torch framework. Key contributions included adding support for various CuDNN functions like convolution and max pooling, and fixing serialization issues.
cudnnffinvidia
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.